The handwriting of Image Segmentation Using the K-Means Clustering Algorithm with Contrast Stretching and Histogram Equalization

Abstract

An analysis of handwritten documents is a scientific technique to understand a writer's personality using handwriting scratches and patterns. Graphologists have identified human characters using visual observations. The identification process requires a long time because the observations are conducted comprehensively and examining one by one of the letters or words of the overall handwriting. Therefore, we need a system that automatically identifies the characters of human personalities based on handwriting, requires a shorter period, and provides objectivity. Handwriting image processing to identify human characters has been developed in various fields, such as education, medicine, psychology, and criminology. In image processing, segmentation is an important stage to separate an object from its background. On the other hand, the k-means clustering algorithm is an algorithm to classify some cluster regions based on certain characteristics. Therefore, it can be implemented at the segmentation stage of handwriting images. This research started with data acquisition. The data employed constituted scans of handwriting obtained from graphologists. Then, the image quality improvement employed contrast stretching and histogram equalization. The next step was image segmentation using the k-means clustering algorithm. Segmentation was conducted by varying k values to gain the best segmentation results. The evaluation was conducted by comparing the results of segmentation images with the results of reference images. The reference images were obtained from segmentation images using Otsu's thresholding method. Otsu's method (1979) has been widely applied in various research on segmentation and produced good accuracy. Therefore, this study applied the image segmentation results with Otsu's method as a reference. The results showed that (1) the highest evaluation indicator was in the segmentation results without pre-processing, and (2) the k value was = 2 with the average accuracy of 100%, the average sensitivity of 100%, and the average specificity of 100%.